[ { "@id": "https://w3id.org/np/RARv-ABCr9uAsxKeqU4U_I7aX2d5v-JtgMOfZatwmlmM8/assertion", "@graph": [ { "@id": "https://w3id.org/np/RARv-ABCr9uAsxKeqU4U_I7aX2d5v-JtgMOfZatwmlmM8/few-shot-eurosat-synthesis", "@type": [ "https://w3id.org/sciencelive/o/terms/Research-Synthesis" ], "http://purl.org/dc/terms/subject": [ { "@id": "http://www.wikidata.org/entity/Q110797734" }, { "@id": "http://www.wikidata.org/entity/Q199687" }, { "@id": "http://www.wikidata.org/entity/Q3001793" }, { "@id": "http://www.wikidata.org/entity/Q6027324" } ], "http://purl.org/spar/cito/isSupportedBy": [ { "@id": "https://w3id.org/sciencelive/np/RA7OZmOmun07jDm8q6lq7Ris1W0MU3rptcp3bphOWUJj8" }, { "@id": "https://w3id.org/sciencelive/np/RA9PP1TVbvRwEv9UNHYfGWvfCXtxCe3f6_xJnf-YbAgSc" }, { "@id": "https://w3id.org/sciencelive/np/RAUS6GbT3Bu-Np0Ue73q58G_c2HilLhh95Y2b8W18o--M" }, { "@id": "https://w3id.org/sciencelive/np/RAqs99x7CDi1tutYKJ6J1zes8PEbUhfjbpf3dkdqSoffQ" } ], "http://schema.org/endDate": [ { "@value": "2026-04-18", "@type": "http://www.w3.org/2001/XMLSchema#date" } ], "http://www.w3.org/2000/01/rdf-schema#label": [ { "@value": "Within-domain few-shot learning is recommended over cross-domain transfer for satellite imagery classification" } ], "https://w3id.org/sciencelive/o/terms/hasConditionsDescription": [ { "@value": "Applicable to optical satellite imagery (Sentinel-2, Landsat) for land cover and habitat classification tasks where labeled examples are scarce. Tested on 10 broad EuroSAT land cover categories at 10 m ground resolution using RGB bands only." } ], "https://w3id.org/sciencelive/o/terms/hasLimitationsDescription": [ { "@value": "Not tested on fine-grained habitat discrimination such as distinguishing vegetation subtypes within Natura 2000 sites. Only RGB bands used — near-infrared and shortwave infrared bands, which are critical for vegetation analysis, were not included. Results may differ for non-optical sensors (SAR, LiDAR) or very high resolution imagery. The within-domain experiment used a different base/novel class split than a real monitoring scenario would require." } ], "https://w3id.org/sciencelive/o/terms/hasRecommendationDescription": [ { "@value": "For Earth observation researchers with limited labeled satellite data: if any labeled satellite imagery is available for your region (even for different classes than your target), use within-domain few-shot learning — train on common classes and classify rare ones. If no satellite training data exists at all, use an off-the-shelf ImageNet-pretrained model as a feature extractor for initial screening, then invest in labeling a small satellite dataset to improve accuracy. Complex meta-learning pipelines are not necessary — standard supervised pretraining achieves comparable results." } ], "https://w3id.org/sciencelive/o/terms/hasSynthesisDescription": [ { "@value": "Four experiments comparing within-domain and cross-domain few-shot learning on Sentinel-2 satellite imagery show that within-domain transfer (training on common satellite land cover classes, classifying rare ones) achieves 82% accuracy with 5 labeled examples per class, while cross-domain transfer from everyday photographs achieves 67–76% depending on backbone architecture and training method. The domain gap between photographs and satellite imagery reduces accuracy by 6–15 percentage points. Supervised pretraining on everyday photographs matches episodic meta-learning with 12 times less training time, but both cross-domain approaches remain below within-domain accuracy." } ] } ] }, { "@id": "https://w3id.org/sciencelive/np/RA5TJVZ0_5Knzxd4OtOoZgO6ZspWHwVCSLWNNd7V9H6QQ/assertion", "@graph": [ { "@id": "https://w3id.org/sciencelive/np/RA5TJVZ0_5Knzxd4OtOoZgO6ZspWHwVCSLWNNd7V9H6QQ/soroye2020-tei-mechanism-replicates-at-fit-but-projection-is-grid-coupled", "@type": [ "https://w3id.org/sciencelive/o/terms/Research-Synthesis" ], "http://purl.org/dc/terms/subject": [ { "@id": "http://www.wikidata.org/entity/Q125928" }, { "@id": "http://www.wikidata.org/entity/Q2922293" }, { "@id": "http://www.wikidata.org/entity/Q5629401" } ], "http://purl.org/spar/cito/isSupportedBy": [ { "@id": "https://w3id.org/sciencelive/np/RAD19jydIHgfVpRQiA8mqvVUefOd7FFwA4tLIfkXmOJmc" }, { "@id": "https://w3id.org/sciencelive/np/RAPZMgcYbScSAXnrnSySQwZzgSA_rn-xodlMxNlwwQYY8" }, { "@id": "https://w3id.org/sciencelive/np/RAa4QR41Hot9zxujcrCyTo82Ij7oaw_6z8zk8NxDqoJFM" } ], "http://schema.org/endDate": [ { "@value": "2026-05-10", "@type": "http://www.w3.org/2001/XMLSchema#date" } ], "http://www.w3.org/2000/01/rdf-schema#label": [ { "@value": "Soroye et al. 2020's TEI-based extirpation mechanism is substrate-robust at fit time but grid-coupled at projection time for low-N species; here is the diagnostic and the recommended reporting protocol" } ], "https://w3id.org/sciencelive/o/terms/hasConditionsDescription": [ { "@value": "- Region: Iberian peninsula (the three synthesised chains all use Iberian Bombus only).\n- Species set: Bombus species observed in GBIF on the Iberian peninsula in 1901–2014 (31 species in the joint analysis).\n- Climate forcing for fit: CRU TS 3.24.01 monthly temperature and precipitation, identical to Soroye et al. 2020 (Figshare deposit).\n- Climate forcing for projection: DestinE Climate DT SSP3-7.0, IFS-NEMO standard, native HEALPix nside=128. Horizons 2020–2029 and 2030–2039 only (DestinE archive populated through 2039 at time of analysis).\n- Spatial substrates analysed: CEA (~100 km), HEALPix-NESTED nside=64 (~92 km), HEALPix-NESTED nside=128 (~46 km) on the WGS84 ellipsoid.\n- GLMM specification: Soroye et al. 2020's full formula with main effects + four predictor interaction terms + per-species random intercept.\n- Inference: full-posterior NUTS via bambi/PyMC; HDIs reported.\n- This synthesis covers BOTH the fit-time substrate-robustness AND the projection-time grid-coupling — these are the two complementary aspects of the same three-chain constellation." } ], "https://w3id.org/sciencelive/o/terms/hasLimitationsDescription": [ { "@value": "1. Three substrates only (CEA, HEALPix nside=64, HEALPix nside=128). Whether the same fit-time substrate-robustness AND projection-time grid-coupling pattern hold at coarser substrates (nside=32 or 16) or against non-HEALPix grids was not directly tested.\n\n2. One region (Iberian peninsula). High-latitude or alpine Bombus systems may behave differently — niche margins, sampling effort, and per-species cell counts all differ.\n\n3. One climate dataset for FIT (CRU TS 3.24.01) and one for PROJECTION (DestinE Climate DT SSP3-7.0, IFS-NEMO standard). Multi-model ensemble robustness and reanalysis-substitution robustness are not addressed.\n\n4. The recommended n_cells ≥ 10 threshold is empirically calibrated on this specific analysis. Whether 10 generalises is open. Below the threshold the per-species ranking is grid-coupled; above it, the recommended protocol yields ρ_Spearman = +0.97 across substrates — but the empirical threshold may be different for taxa with different occupancy distributions.\n\n5. Variant (c) \"shared CEA reference standardisation\" was tested as an approximation of refit-with-shared-standardisation, not as a true refit. The most rigorous test of \"is standardisation reference choice alone sufficient\" is to refit the GLMM with shared (μ, σ) and re-project — a deferred follow-up.\n\n6. Time horizon. SSP3-7.0 mid-term (2030–2039) is the strongest test horizon currently available. End-of-century horizons (2046–2055, 2076–2085) are deferred until the DestinE Climate DT archive extends past 2050.\n\n7. The synthesis treats the mechanism's projection-time grid-coupling as a methodological caveat, not a refutation of Soroye et al.'s biological claim. This framing assumes the user accepts the mechanism's epistemic separation between FIT (substrate-robust) and PROJECTION (qualified).\n" } ], "https://w3id.org/sciencelive/o/terms/hasRecommendationDescription": [ { "@value": "1. When replicating Soroye-style TEI extirpation models, expect substrate-robust headline coefficients within ±30% across factor-of-2 HEALPix resolution changes — substrate sensitivity at FIT time is small. Cite multiple substrates jointly when claiming substrate-robustness; a single substrate is not sufficient evidence.\n\n2. When projecting to future climate, ALWAYS filter per-species reporting to species with at least 10 occupied + active cells per substrate. Below this, per-species ranking is grid-coupled and not interpretable.\n\n3. At projection time, drop the GLMM interaction terms from the linear predictor used for extrapolation. Keep them in the FIT (they are part of Soroye's specification) but use main-effects-only η to project. This single change lifts cross-substrate Spearman ρ from +0.59 (full GLMM) to +0.97 at n ≥ 10.\n\n4. Cross-check ANY GLMM-based ranking against the substrate-invariant physical metric \"mean future TEI per species\" or \"fraction of cells with TEI_future > 0.5\". If the GLMM main-effects ranking and the physical ranking disagree for a given species, the GLMM is unreliable for that species at that substrate.\n\n5. CiTO encoding: cito:confirms for the fit-time substrate-robustness; cito:qualifies for projection-time use without the recommended protocol; cito:extends to this diagnostic when reporting projection results at a new substrate pair.\n\n6. The recommended protocol is empirically calibrated against HEALPix nside=64 vs nside=128 for Iberian Bombus on DestinE Climate DT SSP3-7.0. Different taxon/region/forcing combinations may require different thresholds; the protocol is a recommendation, not a hard constraint." } ], "https://w3id.org/sciencelive/o/terms/hasSynthesisDescription": [ { "@value": "Synthesising the three sibling FORRT chains on Iberian Bombus (canonical CEA + HEALPix nside=64 replication, HEALPix nside=128 substrate extension, and this cross-substrate diagnostic), Soroye et al.'s (2020) TEI-based extirpation mechanism resolves into two empirically distinct claims:\n\n(A) At fit time the mechanism IS substrate-robust on Iberian Bombus. The GLMM coefficient on standardised TEI_delta is positive and credibly above zero at three independent pixelisations (CEA +0.479; HEALPix nside=64 +0.454; HEALPix nside=128 +0.347), with all three estimates within ±30 percent. Soroye's central biological claim REPLICATES.\n\n(B) At projection time the per-species ranking is NOT substrate-robust below ~10 occupied + active cells per substrate. Per-species community-mean η differs by 1 to 9 logits between nside=64 and nside=128 under DestinE Climate DT SSP3-7.0 for the 18 of 31 species below the cell-count threshold, including narrowly-distributed Pyrenean specialists (B. pyrenaeus, B. mucidus, B. mendax). The substrate-coupling is mechanistically diagnosable: the GLMM interaction term sc_TEI_delta:sc_PEI_delta compounds substrate-specific predictor standardisation quadratically when future predictors extrapolate 2–4σ outside the training distribution.\n\nThese two claims are not in tension — they describe different parts of the modelling pipeline. The original mechanism is sound; the projection-time application is qualified. Three principled fixes are empirically validated: (i) per-species cell-count filter at n_cells ≥ 10; (ii) drop the GLMM interaction terms at projection time only; (iii) cross-check against substrate-invariant physical metrics (mean future TEI, fraction TEI_future > 0.5). At n ≥ 10 with main-effects-only η, cross-substrate Spearman ρ = +0.97 (mid-term horizon) — substrate-stable." } ] } ] }, { "@id": "https://w3id.org/sciencelive/np/RAcDYOu65z09jUbDwd_c2OxGI9KUPZmLszxUlLVOyzt3M/assertion", "@graph": [ { "@id": "https://w3id.org/sciencelive/np/RAcDYOu65z09jUbDwd_c2OxGI9KUPZmLszxUlLVOyzt3M/mediterranean-thermal-niche-exceedance-constellation-iberia", "@type": [ "https://w3id.org/sciencelive/o/terms/Research-Synthesis" ], "http://purl.org/dc/terms/subject": [ { "@id": "http://www.wikidata.org/entity/Q107482001" }, { "@id": "http://www.wikidata.org/entity/Q125928" }, { "@id": "http://www.wikidata.org/entity/Q12837" }, { "@id": "http://www.wikidata.org/entity/Q139684181" }, { "@id": "http://www.wikidata.org/entity/Q25407" }, { "@id": "http://www.wikidata.org/entity/Q27532" }, { "@id": "http://www.wikidata.org/entity/Q5629401" }, { "@id": "http://www.wikidata.org/entity/Q7150" } ], "http://purl.org/spar/cito/isSupportedBy": [ { "@id": "https://w3id.org/sciencelive/np/RA-GY814xxcpEsUWozEJKHGG39bDV8gkbor7OhX8QpVPE" }, { "@id": "https://w3id.org/sciencelive/np/RA1q6c0fG2bMbiozF8Az2UpIfzAzqp8hoVEl6QIzfUpH8" }, { "@id": "https://w3id.org/sciencelive/np/RA5TJVZ0_5Knzxd4OtOoZgO6ZspWHwVCSLWNNd7V9H6QQ" }, { "@id": "https://w3id.org/sciencelive/np/RAKH9XeZn3CUr9WaFKMC3O2pT_HJJ96c3jTa6v6dWEE3c" }, { "@id": "https://w3id.org/sciencelive/np/RAPw01nGqrY3V9ech7nd3gUOyoW52N6SspeKTa41neLBo" }, { "@id": "https://w3id.org/sciencelive/np/RAfdV1yB1JksVJ7dJYwECRHVMhNzbGcjUAa6UreqG_fM4" }, { "@id": "https://w3id.org/sciencelive/np/RAzhqiFNu8xIORz_w-sq3Ip8aXH8yvokLt1qaOtXNcu-k" } ], "http://schema.org/endDate": [ { "@value": "2026-05-15", "@type": "http://www.w3.org/2001/XMLSchema#date" } ], "http://www.w3.org/2000/01/rdf-schema#label": [ { "@value": "Iberian thermal-niche-exceedance constellation — Bombus (Soroye 2020) + Lacertidae (Sinervo 2010) under Destination Earth Climate DT SSP3-7.0" } ], "https://w3id.org/sciencelive/o/terms/hasConditionsDescription": [ { "@value": "Region: Iberian Peninsula (Spain, Portugal, Andorra, Gibraltar — bbox roughly -10°W to 4°E, 35°N to 44°N).\n\nClimate forcing: Destination Earth Climate Digital Twin SSP3-7.0 IFS-NEMO at HEALPix nside=128 native, 2020-2039 archive coverage. Single climate realisation (no ensemble). Daily-aggregated from 4-times-daily 2m temperature snapshots.\n\nTaxa: heliothermic / diurnal-active ectotherms (Lacertidae lizards via Sinervo 2010) and warm-adapted social insects (Bombus pollinators via Soroye 2020). The mechanism class (thermal-niche exceedance under climate warming as a driver of local extirpation/extinction) is conserved across the two replications; the operational instantiation differs (h_r threshold for Lacertidae, TEI-delta GLMM coefficient for Bombus).\n\nMethods: deterministic h_r threshold comparison (Lacertidae chain) and Bayesian GLMM TEI fit (Bombus chain). Both implemented in Python (xarray + healpix-geo + cfgrib + pygbif) with Snakefile pipelines, MIT-licensed and Zenodo-archived. Both follow the same FORRT chain shape (Quote → AIDA → Claim → Replication Study → Replication Outcome → CiTO Citation + Research Software).\n\nTime period of validity: 2020-2039 only for the projection-horizon component; 2026-05 for the FAIR-archive / nanopub-network state. Future DestinE archive expansion may enable testing of Sinervo's 2050 / 2080 horizons.\n\nPinning: HEALPix NESTED ordering throughout; geographic data uses healpix-geo (NOT healpy, per DOMAIN.md); arviz<0.22 for PyMC compatibility; numpy 2.x; pymc 5.25+." } ], "https://w3id.org/sciencelive/o/terms/hasLimitationsDescription": [ { "@value": "Five limitations bounding the synthesis statement, ordered by interpretive importance:\n\n(1) Single climate realisation. Both chains used a single DestinE Climate DT SSP3-7.0 IFS-NEMO realisation. Interannual variability in extreme-heat-day frequency (which dominates the Lacertidae signal under S3a) cannot be separated from forced trend on a single realisation. Cross-realisation ensemble would refine the constellation's quantitative statements about decadal direction and tail behaviour.\n\n(2) Cross-taxon generalisation. Two taxa (Bombus, Lacertidae) is a small cross-taxon sample. The constellation conclusion (\"mechanism class transfers but predictions are prior-conditional\") is supported by these two cases but should be re-tested as further taxa join the constellation. Candidate next-taxon chains: amphibians (Iberian Salamandridae / Discoglossidae), other diurnal heliotherms (Iberian Anguidae if their physiology is in scope), other warm-adapted insects (Iberian Apidae beyond Bombus).\n\n(3) Family-level priors. Both replications used family-level priors as the baseline. The Lacertidae chain explicitly flagged species-specific T_b (Iberian Podarcis / Iberolacerta / Timon) as deferred to a future iteration; the Bombus chain similarly used family-level GLMM coefficients rather than species-specific refits. Constellation-level conclusions about prior-conditional behaviour would tighten under species-specific refinement.\n\n(4) DestinE-horizon limit. The 2020-2039 archive coverage does not extend to Sinervo's 2050 / 2080 headline-year projections nor to the multi-decadal climate-trajectory regimes where both source papers' mechanisms ramp up. The constellation's statement that \"mechanism does not trigger at near-term horizons\" is honest for the reachable archive but does not preclude triggering at longer horizons. Re-running the constellation against an expanded DestinE archive when 2050+ becomes available is the natural follow-up.\n\n(5) Sub-daily climate-twin access. Sinervo SOM Equation S2 is operationally daily-Tmax-based, but the mechanism's underlying physiology is intra-day (basking-window thermal restriction). The Polytope sub-daily probe across both chains returned \"credentials absent — fallback unused\" status; the sinusoidal Tmin-Tmax diurnal-cycle reconstruction was scoped but not exercised. If sub-daily access becomes available, the lizards mechanism may activate at lower compound-prior thresholds than reported here." } ], "https://w3id.org/sciencelive/o/terms/hasRecommendationDescription": [ { "@value": "For practitioners testing thermal-niche-exceedance mechanisms in new taxa or under new climate-digital-twin archives:\n\n(1) Report sensitivity matrices, not single-point headline figures. A 2 x 3 cross-product of T_b prior and reproductive-window choice (or analogous parameter axes) surfaces prior-conditional behaviour that single-figure headlines hide. This applies even when the baseline result is zero — the value of the matrix is showing the boundary of the parameter space where the mechanism activates.\n\n(2) Run a substrate-sensitivity diagnostic at two HEALPix resolutions whenever per-species rankings are reported. Use the NESTED bit-shift parent = pix >> 2 for the downsample (no re-projection, free of regridding error). Report Spearman rho separately for well-sampled species (n_cells >= 10) and the low-N subset; the latter is expected to be grid-coupled per Lobo 2007 / Hurlbert & Jetz 2007, but the diagnostic confirms it operationally for the specific dataset.\n\n(3) Honestly scope CiTO citation type to the projection horizon. Use `cito:extends` (not `cito:confirms` / `disputes`) when the digital-twin archive does not cover the source paper's projection horizon. The mechanism's behaviour at reachable horizons is a separate Outcome from the source paper's headline projection; conflating them produces over-claiming.\n\n(4) Publish a cross-taxon constellation Synthesis whenever two or more chains share a mechanism class and a regional / temporal scope. The constellation-level statement (this nanopub) is qualitatively different from any single chain's Outcome, and the cross-chain consistency or divergence is itself a finding.\n\n(5) Mirror prior chains' substrate, env, and pinning conventions wherever possible. The Bombus constellation established HEALPix-NESTED nside=128 as the DestinE-native substrate and a specific environment.yml pinning philosophy (lower-bounds + arviz<0.22 ceiling). The lizards chain reused both with no modification, which made the cross-replication composition trivial — the constellation Synthesis benefits operationally from this consistency." } ], "https://w3id.org/sciencelive/o/terms/hasSynthesisDescription": [ { "@value": "Thermal-niche-exceedance replicates as a class of mechanism across pollinators and ectotherms in Iberia, but operational extinction projections depend on parameter choices and time horizon in ways the source papers' headline figures do not surface. The two-taxon constellation shows: (i) Soroye 2020's TEI-based extirpation mechanism for Iberian Bombus is substrate-robust at fit time (GLMM coefficient sc_TEI_delta = +0.454 at HEALPix nside=64, +0.347 at nside=128 — both within ~30 % of the published continental +0.479) but grid-coupled at projection time for low-N species, surfaced operationally by the substrate-sensitivity sibling chain; (ii) Sinervo 2010's h_r mechanism for Iberian Lacertidae predicts zero local extinction at DestinE-reachable horizons (2020-2039) under family-mean priors, and only ~3 % at compound worst-plausible priors (Iberolacerta T_b + May-June reproductive window), with the per-species substrate-sensitivity diagnostic confirming substrate-robust rankings for well-sampled species (Spearman rho = 0.951 for n_cells >= 10). Across both taxa, three constellation-level claims emerge: (a) the mechanism class transfers cross-taxon when properly operationalised, but operational predictions are prior-conditional in ways the original papers' single-figure headlines do not surface; (b) DestinE-reachable horizons (2020-2039) sit below the activation thresholds for the lizards mechanism, so Sinervo's 24 % / 46 % / 2050 / 2080 projections cannot be tested under current archive coverage — the gap between near-term-DestinE and source-paper-projection horizons is itself a finding about the operational reach of climate-digital-twin replications; (c) the per-species rare-species-ranking grid-coupling caveat (Lobo et al. 2007; Hurlbert & Jetz 2007) manifests differently across taxa — as ranking-shuffle at non-zero rates in the Bombus chain, as signal-collapse below threshold in the Lacertidae chain — both consistent with the established literature, neither novel." } ] } ] }, { "@id": "https://w3id.org/sciencelive/np/RA6r8sefdZHemsSWVZVo7nXdvydNqCm-VfrQpnmmRBrfA/assertion", "@graph": [ { "@id": "https://w3id.org/sciencelive/np/RA6r8sefdZHemsSWVZVo7nXdvydNqCm-VfrQpnmmRBrfA/spherical-ml-substrate-synthesis-2026", "@type": [ "https://w3id.org/sciencelive/o/terms/Research-Synthesis" ], "http://purl.org/dc/terms/subject": [ { "@id": "http://www.wikidata.org/entity/Q1507383" }, { "@id": "http://www.wikidata.org/entity/Q2539" }, { "@id": "http://www.wikidata.org/entity/Q47041" }, { "@id": "http://www.wikidata.org/entity/Q5629401" }, { "@id": "http://www.wikidata.org/entity/Q56321065" } ], "http://purl.org/spar/cito/isSupportedBy": [ { "@id": "https://w3id.org/sciencelive/np/RA0TakYbwjs9vdc2AXyKxaCj54u5vr8zIrNhebEEskWRc" }, { "@id": "https://w3id.org/sciencelive/np/RAoW3q1q1Wyt5DXbFl2PI3woyhuYZuU8HYtJ3m0LyrP9M" }, { "@id": "https://w3id.org/sciencelive/np/RAvIzcWGL89mxdBXTTjgRRd0QJBBAdu7wUqkdHRCssSqs" }, { "@id": "https://w3id.org/sciencelive/np/RAydqzcPo3ZNMYU2Gk9wd4u4OgITCaXwL01IQtxyoBloA" } ], "http://schema.org/endDate": [ { "@value": "2026-05-07", "@type": "http://www.w3.org/2001/XMLSchema#date" } ], "http://www.w3.org/2000/01/rdf-schema#label": [ { "@value": "The HEALPix-NESTED substrate makes sphere-aware ML latitude-invariant, discipline-transferable, and biodiversity-attribution-ready" } ], "https://w3id.org/sciencelive/o/terms/hasConditionsDescription": [ { "@value": "Scope: global ML detection / classification tasks where features can be expressed on a HEALPix-NESTED grid, including but not limited to sea-surface-temperature anomaly fields (NOAA OISST v2.1, Copernicus Marine SST, ERA5), tropospheric or stratospheric atmospheric fields (DLWP-HEALPix forecast outputs, ClimateNet), marine biodiversity occurrences (GBIF, OBIS) at coarsest-feature resolution, and synthetic Gaussian-random-field samples with compact features. Methods: sphere-harmonic transforms via healpy.map2alm / alm2map, sphere-harmonic-domain convolutions via aₗₘ → aₗₘ · fₗ · bₗ, equal-area cell aggregation via numpy.bincount on healpy.ang2pix(..., nest=True). Domains: cosmology, climate, Earth observation, marine biodiversity, atmospheric science. The latitude-invariance and cross-discipline-transfer claims hold for any compact-feature detection task at fixed angular scale on the HEALPix substrate; the biodiversity-attribution claim is documented for the 2011 Ningaloo Niño event in the Western Australian region but the substrate-and-method combination generalises to any documented MHW or atmospheric-event case with available occurrence data." } ], "https://w3id.org/sciencelive/o/terms/hasLimitationsDescription": [ { "@value": "(1) The within-discipline and cross-discipline tests use synthetic data with controlled feature physics; the substrate-dependence is demonstrated via a minimal (max, mean, std) matched-filter feature triple to isolate the substrate effect from the model class, not via a deep learned representation. The numerical magnitudes (1.000 vs 0.500, 1.000 vs 0.845) reflect the geometric mechanism cleanly but are upper bounds on what richer learned representations like DeepSphere graph CNNs or foscat scattering networks would deliver on real data. \n(2) The real-data biodiversity-attribution case (chain B) uses a 3-year-baseline (2008–2010) simplification of the canonical Hobday et al. 2016 30-year 1991–2020 climatology and a fixed +1.5 °C anomaly threshold rather than the per-day-of-year 90th-percentile threshold; the qualitative spatial footprint matches the documented Ningaloo Niño but the per-cell MHW-day count is approximate. \n(3) The biodiversity-overlap statistic is exposure-only (records on MHW cells), not a causal attribution of biodiversity change to MHW conditions; the linkage to Wernberg et al. 2016 kelp regime shift is via spatial-temporal coincidence with the documented event. \n(4) The cross-discipline transfer test uses synthetic discipline regimes constructed to share feature physics across different background spectra; true cross-discipline transfer from real cosmology data (e.g., Planck CMB on HEALPix) to real climate data (e.g., DLWP-HEALPix forecasts on HEALPix) would require integrating with foscat or DeepSphere as future work. \n(5) All HEALPix work in this synthesis uses NESTED ordering throughout per the project-wide convention; results for RING-ordered HEALPix at the same resolution are expected to be equivalent but were not separately tested." } ], "https://w3id.org/sciencelive/o/terms/hasRecommendationDescription": [ { "@value": "(1) When training ML detection or classification models on globally-distributed Earth-observation, climate, or biodiversity data, render features on a HEALPix-NESTED substrate (spherical for cosmology / synthetic experiments, WGS84-ellipsoidal via healpix-geo for geoscience) before applying convolutions; the substrate choice is what separates \"works at the equator only\" from \"works at every latitude\". \n(2) When integrating biodiversity occurrence data with Copernicus Marine SST, NOAA OISST, ERA5, or DestinE climate-model output for impact-attribution work, co-locate all four sources on a shared HEALPix-NESTED grid (NESTED ordering for hierarchical bit-shift refinement) at the resolution that matches the coarsest input, then perform overlap statistics there; do not perform per-source lat-lon aggregation followed by raster-level joins. (3) When evaluating sphere-aware versus flat ML pipelines, report accuracy at multiple test latitude bands and on at least one cross-discipline transfer regime; in-distribution-only metrics under-state the substrate effect. (4) Investments in sphere-aware models from one discipline (cosmology DeepSphere, foscat scattering networks, DLWP-HEALPix global weather forecasting) carry directly over to the other disciplines on the same HEALPix substrate; budget integration work as a feature-extractor port rather than a from-scratch retrain." } ], "https://w3id.org/sciencelive/o/terms/hasSynthesisDescription": [ { "@value": "Three independent tests on the HEALPix-NESTED substrate jointly establish that sphere-aware operators recover detection accuracy lat-lon flat operators lose, transfer across discipline pairs without retraining, and integrate with marine biodiversity occurrence data on a single shared substrate — without re-projection at any step. The within-discipline test (chain A, notebook 04) shows the lat-lon-flat matched filter collapsing from 1.000 to 0.500 chance at 70–80° latitude while the sphere-harmonic band-pass matched filter holds at 1.000 across all four test bands. The cross-discipline test (chain C, notebook 06) shows the same sphere-aware pipeline transferring at 1.000 from a cosmology-like training domain to a climate-like test domain without retraining, while the flat baseline drops to 0.845 on the same transfer. The real-data test (chain B, notebook 05) shows that when the climate-event field and the biodiversity occurrence field meet on the same HEALPix substrate, 94.0 percent of 765 marine GBIF records during the documented 2011 Ningaloo Niño event sat on cells that experienced marine-heatwave conditions in the same window Wernberg et al. 2016 documented the kelp regime shift in. The shared property — that sphere-harmonic convolution is exactly rotation-equivariant on the sphere, while lat-lon convolution is only translation-equivariant in pixel space — is what makes the substrate the right common DGGS for Copernicus EO, Destination Earth climate models, and biodiversity-impact attribution to interoperate." } ] } ] } ]